This article estimates the effect of sex education in a large-scale cluster randomized trial covering a third of Botswana that varied both the message and the messenger. Educational messages delivered by near-peers led to a statistically significant 40% reduction in adolescent pregnancy incidence, whereas the same messages delivered by government teachers showed no detectable effect. While both types of messengers successfully change students’ beliefs, students appear to be persuaded by near-peers to change their behavior but are not persuaded by teachers. These results demonstrate the first-order role messengers can play in influencing behavior and motivate greater use of near-peer messengers to deliver sex education at scale.